Line data Source code
1 : /* SPDX-License-Identifier: LGPL-2.1+ */
2 :
3 : #include <linux/bpf_insn.h>
4 : #include <string.h>
5 : #include <sys/mman.h>
6 : #include <unistd.h>
7 :
8 : #include "bpf-firewall.h"
9 : #include "bpf-program.h"
10 : #include "load-fragment.h"
11 : #include "manager.h"
12 : #include "missing.h"
13 : #include "rm-rf.h"
14 : #include "service.h"
15 : #include "test-helper.h"
16 : #include "tests.h"
17 : #include "unit.h"
18 : #include "virt.h"
19 :
20 : /* We use the same limit here that PID 1 bumps RLIMIT_MEMLOCK to if it can */
21 : #define CAN_MEMLOCK_SIZE (64U*1024U*1024U)
22 :
23 1 : static bool can_memlock(void) {
24 : void *p;
25 : bool b;
26 :
27 : /* Let's see if we can mlock() a larger blob of memory. BPF programs are charged against
28 : * RLIMIT_MEMLOCK, hence let's first make sure we can lock memory at all, and skip the test if we
29 : * cannot. Why not check RLIMIT_MEMLOCK explicitly? Because in container environments the
30 : * RLIMIT_MEMLOCK value we see might not match the RLIMIT_MEMLOCK value actually in effect. */
31 :
32 1 : p = mmap(NULL, CAN_MEMLOCK_SIZE, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_SHARED, -1, 0);
33 1 : if (p == MAP_FAILED)
34 0 : return false;
35 :
36 1 : b = mlock(p, CAN_MEMLOCK_SIZE) >= 0;
37 1 : if (b)
38 0 : assert_se(munlock(p, CAN_MEMLOCK_SIZE) >= 0);
39 :
40 1 : assert_se(munmap(p, CAN_MEMLOCK_SIZE) >= 0);
41 1 : return b;
42 : }
43 :
44 1 : int main(int argc, char *argv[]) {
45 1 : struct bpf_insn exit_insn[] = {
46 : BPF_MOV64_IMM(BPF_REG_0, 0), /* drop */
47 : BPF_EXIT_INSN()
48 : };
49 :
50 1 : _cleanup_(rm_rf_physical_and_freep) char *runtime_dir = NULL;
51 1 : CGroupContext *cc = NULL;
52 1 : _cleanup_(bpf_program_unrefp) BPFProgram *p = NULL;
53 1 : _cleanup_(manager_freep) Manager *m = NULL;
54 : Unit *u;
55 : char log_buf[65535];
56 : struct rlimit rl;
57 : int r;
58 : union bpf_attr attr;
59 1 : bool test_custom_filter = false;
60 1 : const char *test_prog = "/sys/fs/bpf/test-dropper";
61 :
62 1 : test_setup_logging(LOG_DEBUG);
63 :
64 1 : if (detect_container() > 0)
65 0 : return log_tests_skipped("test-bpf fails inside LXC and Docker containers: https://github.com/systemd/systemd/issues/9666");
66 :
67 1 : assert_se(getrlimit(RLIMIT_MEMLOCK, &rl) >= 0);
68 1 : rl.rlim_cur = rl.rlim_max = MAX3(rl.rlim_cur, rl.rlim_max, CAN_MEMLOCK_SIZE);
69 1 : (void) setrlimit(RLIMIT_MEMLOCK, &rl);
70 :
71 1 : if (!can_memlock())
72 1 : return log_tests_skipped("Can't use mlock(), skipping.");
73 :
74 0 : r = enter_cgroup_subroot();
75 0 : if (r == -ENOMEDIUM)
76 0 : return log_tests_skipped("cgroupfs not available");
77 :
78 0 : assert_se(set_unit_path(get_testdata_dir()) >= 0);
79 0 : assert_se(runtime_dir = setup_fake_runtime_dir());
80 :
81 0 : r = bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &p);
82 0 : assert(r == 0);
83 :
84 0 : r = bpf_program_add_instructions(p, exit_insn, ELEMENTSOF(exit_insn));
85 0 : assert(r == 0);
86 :
87 0 : if (getuid() != 0)
88 0 : return log_tests_skipped("not running as root");
89 :
90 0 : r = bpf_firewall_supported();
91 0 : if (r == BPF_FIREWALL_UNSUPPORTED)
92 0 : return log_tests_skipped("BPF firewalling not supported");
93 0 : assert_se(r > 0);
94 :
95 0 : if (r == BPF_FIREWALL_SUPPORTED_WITH_MULTI) {
96 0 : log_notice("BPF firewalling with BPF_F_ALLOW_MULTI supported. Yay!");
97 0 : test_custom_filter = true;
98 : } else
99 0 : log_notice("BPF firewalling (though without BPF_F_ALLOW_MULTI) supported. Good.");
100 :
101 0 : r = bpf_program_load_kernel(p, log_buf, ELEMENTSOF(log_buf));
102 0 : assert(r >= 0);
103 :
104 0 : if (test_custom_filter) {
105 0 : attr = (union bpf_attr) {
106 0 : .pathname = PTR_TO_UINT64(test_prog),
107 0 : .bpf_fd = p->kernel_fd,
108 : .file_flags = 0,
109 : };
110 :
111 0 : (void) unlink(test_prog);
112 :
113 0 : r = bpf(BPF_OBJ_PIN, &attr, sizeof(attr));
114 0 : if (r < 0) {
115 0 : log_warning_errno(errno, "BPF object pinning failed, will not run custom filter test: %m");
116 0 : test_custom_filter = false;
117 : }
118 : }
119 :
120 0 : p = bpf_program_unref(p);
121 :
122 : /* The simple tests succeeded. Now let's try full unit-based use-case. */
123 :
124 0 : assert_se(manager_new(UNIT_FILE_USER, MANAGER_TEST_RUN_BASIC, &m) >= 0);
125 0 : assert_se(manager_startup(m, NULL, NULL) >= 0);
126 :
127 0 : assert_se(u = unit_new(m, sizeof(Service)));
128 0 : assert_se(unit_add_name(u, "foo.service") == 0);
129 0 : assert_se(cc = unit_get_cgroup_context(u));
130 0 : u->perpetual = true;
131 :
132 0 : cc->ip_accounting = true;
133 :
134 0 : assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressAllow", 0, "10.0.1.0/24", &cc->ip_address_allow, NULL) == 0);
135 0 : assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressAllow", 0, "127.0.0.2", &cc->ip_address_allow, NULL) == 0);
136 0 : assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "127.0.0.3", &cc->ip_address_deny, NULL) == 0);
137 0 : assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "10.0.3.2/24", &cc->ip_address_deny, NULL) == 0);
138 0 : assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "127.0.0.1/25", &cc->ip_address_deny, NULL) == 0);
139 0 : assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "127.0.0.4", &cc->ip_address_deny, NULL) == 0);
140 :
141 0 : assert(cc->ip_address_allow);
142 0 : assert(cc->ip_address_allow->items_next);
143 0 : assert(!cc->ip_address_allow->items_next->items_next);
144 :
145 : /* The deny list is defined redundantly, let's ensure it got properly reduced */
146 0 : assert(cc->ip_address_deny);
147 0 : assert(cc->ip_address_deny->items_next);
148 0 : assert(!cc->ip_address_deny->items_next->items_next);
149 :
150 0 : assert_se(config_parse_exec(u->id, "filename", 1, "Service", 1, "ExecStart", SERVICE_EXEC_START, "/bin/ping -c 1 127.0.0.2 -W 5", SERVICE(u)->exec_command, u) == 0);
151 0 : assert_se(config_parse_exec(u->id, "filename", 1, "Service", 1, "ExecStart", SERVICE_EXEC_START, "/bin/ping -c 1 127.0.0.3 -W 5", SERVICE(u)->exec_command, u) == 0);
152 :
153 0 : assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]);
154 0 : assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next);
155 0 : assert_se(!SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next->command_next);
156 :
157 0 : SERVICE(u)->type = SERVICE_ONESHOT;
158 0 : u->load_state = UNIT_LOADED;
159 :
160 0 : unit_dump(u, stdout, NULL);
161 :
162 0 : r = bpf_firewall_compile(u);
163 0 : if (IN_SET(r, -ENOTTY, -ENOSYS, -EPERM))
164 0 : return log_tests_skipped("Kernel doesn't support the necessary bpf bits (masked out via seccomp?)");
165 0 : assert_se(r >= 0);
166 :
167 0 : assert(u->ip_bpf_ingress);
168 0 : assert(u->ip_bpf_egress);
169 :
170 0 : r = bpf_program_load_kernel(u->ip_bpf_ingress, log_buf, ELEMENTSOF(log_buf));
171 :
172 0 : log_notice("log:");
173 0 : log_notice("-------");
174 0 : log_notice("%s", log_buf);
175 0 : log_notice("-------");
176 :
177 0 : assert(r >= 0);
178 :
179 0 : r = bpf_program_load_kernel(u->ip_bpf_egress, log_buf, ELEMENTSOF(log_buf));
180 :
181 0 : log_notice("log:");
182 0 : log_notice("-------");
183 0 : log_notice("%s", log_buf);
184 0 : log_notice("-------");
185 :
186 0 : assert(r >= 0);
187 :
188 0 : assert_se(unit_start(u) >= 0);
189 :
190 0 : while (!IN_SET(SERVICE(u)->state, SERVICE_DEAD, SERVICE_FAILED))
191 0 : assert_se(sd_event_run(m->event, UINT64_MAX) >= 0);
192 :
193 0 : assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->exec_status.code == CLD_EXITED &&
194 : SERVICE(u)->exec_command[SERVICE_EXEC_START]->exec_status.status == EXIT_SUCCESS);
195 :
196 0 : assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next->exec_status.code != CLD_EXITED ||
197 : SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next->exec_status.status != EXIT_SUCCESS);
198 :
199 0 : if (test_custom_filter) {
200 0 : assert_se(u = unit_new(m, sizeof(Service)));
201 0 : assert_se(unit_add_name(u, "custom-filter.service") == 0);
202 0 : assert_se(cc = unit_get_cgroup_context(u));
203 0 : u->perpetual = true;
204 :
205 0 : cc->ip_accounting = true;
206 :
207 0 : assert_se(config_parse_ip_filter_bpf_progs(u->id, "filename", 1, "Service", 1, "IPIngressFilterPath", 0, test_prog, &cc->ip_filters_ingress, u) == 0);
208 0 : assert_se(config_parse_exec(u->id, "filename", 1, "Service", 1, "ExecStart", SERVICE_EXEC_START, "-/bin/ping -c 1 127.0.0.1 -W 5", SERVICE(u)->exec_command, u) == 0);
209 :
210 0 : SERVICE(u)->type = SERVICE_ONESHOT;
211 0 : u->load_state = UNIT_LOADED;
212 :
213 0 : assert_se(unit_start(u) >= 0);
214 :
215 0 : while (!IN_SET(SERVICE(u)->state, SERVICE_DEAD, SERVICE_FAILED))
216 0 : assert_se(sd_event_run(m->event, UINT64_MAX) >= 0);
217 :
218 0 : assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->exec_status.code != CLD_EXITED ||
219 : SERVICE(u)->exec_command[SERVICE_EXEC_START]->exec_status.status != EXIT_SUCCESS);
220 :
221 0 : (void) unlink(test_prog);
222 0 : assert_se(SERVICE(u)->state == SERVICE_DEAD);
223 : }
224 :
225 0 : return 0;
226 : }
|
